QUARTERLY OF APPLIED MATHEMATICS
13$
APRIL 1980
ON CONTINUOUS DEPENDENCE IN FINITE ELASTICITY*
By SCOTT J. SPECTOR (University of Tennessee)
Abstract. We investigate the relation between stability and continuous dependence
for a nonlinearly elastic body at equilibrium. We show that solutions of the governing
equations that lie in a convex, stable set of deformations depend continuously on the
body forces and the surface tractions. The definition of stability used is essentially due to
Hadamard.
1. Introduction. In this note we consider the relation between stability and continuous dependence for a nonlinearly elastic body at equilibrium. We show that solutions of
the governing equations that lie in a convex, uniformly Hadamard-stable set of deformations depend continuously on the body forces and surface tractions.
Gurtin and Spector [1] have shown that Hadamard stability is also sufficient for
uniqueness of solutions of the governing equations. Thus uniqueness and continuous
dependence follow from the same assumptions.
Hadamard stability of a set Q of deformations is the requirement that for some a > 0
f Vu ■/4(V/)V«> a||V«|L2 2(&)
J n*
for every deformation / in Q and every variation u. Here A is the elasticity tensor, the
derivative of the stress response function with respect to the deformation gradient, while
96 is the region of space occupied by the body in a fixed reference configuration.
2. The response function. Stability.
We consider a body 96 and we identify 96 with
the properly regular1 region of R3 it occupies in a fixed reference configuration.
Further,
we denote by 3 and Sf complementary subsets of the boundary d96 with 3 non-empty
and relatively open.
A deformation / (of 96) will be a member of the space
Def = {/e C\9I, IR3):det V/> 0},
while a variation u will be a smooth function u: 96 -* R3 which satisfies
u = 0 on 3.
Here det is the determinant and V the gradient operator in [R3.In particular, V/ is the
tensor field with components (V/% = df/dxj.
* Received February 26, 1979. The author especially wishes to thank M. E. Gurtin for his comments on a
previous version of this manuscript. Thanks are also due to M. Aron and J. J. Roseman for correspondence
concerning continuous dependence.
1 Cf. Fichera [2], p. 351. In particular, St is compact and has piecewise smooth (C1) boundary.
136
SCOTT J. SPECTOR
We assume that the body is elastic with smooth response function S. Thus
S(Vf(x),x)
gives the (Piola-Kirchhoff) stress at any point xel
when the body is deformed by f
Writing F for V/(x) we define the elasticity tensor A(F, x): Lin -<■Lin by2
A(F,x) = dFS(F,x).
(1)
For convenience we write S(Vf) and A(Vf) for the fields on 0$ with values S(Vf(x), x)
and A(Vf(x), x) respectively.
Definition [1], A set Q <=Def is (uniformly) Hadamard-stable
some a > 0
f
or simply H-stable if for
Vu ■A(Vf)Vu> a||Vu||2,w
(2)
for all f e Q and variations u, in which case we write stab(Q) for the largest a with this
property.
In the above inequality
||Vw||hm =| * Vm• Vw
where for any T, V e Lin,
V ' 7 = £ Vij'I'ij.
i.j
We will also have occasion to use the L2-norm of vector fields over 3d and y;
\\U IL2(m■
=
Jn
U ■ U,
\\u\\2L2(y) =|
Jy
U ■ U.
Remark. The above notion of stability is essentially due to Hadamard3 [3, p. 252]. For
a further discussion of stability cf. [1] and [4].
Gurtin and Spector [1] have shown that there is a neighborhood of the reference
configuration which is uniformly H-stable if either:
(a) the reference configuration is positive and natural4; or
(b) S> —d3d and the reference configuration is homogeneous and strongly elliptic.
3. The mixed problem. Continuous dependence. The mixed problem (with dead loading) consists in finding a deformation/that
satisfies:
(i) the equation of virtual work
f S(V/) -Wu=\s-u+\b-u
J<a
(3)
for all variations u; and
2 We use Lin to denote the space of linear transformations from R3 into R3.
3 Hadamard stability is a static "criterion" for stability and its precise relation to dynamic stability is unclear.
Eq. (2) cannot hold for all deformations, since global uniqueness would then follow.
4 A natural configuration whose elasticity tensor is positive definite when restricted to symmetric tensors.
NOTES
137
(ii) the displacement boundary condition
f = d on 3>.
(4)
Here, d e C°(S(, 1R3)is the surface deformation, s e
IR3) the surface traction, and
be
R3) the body force. The triplet (d, s, b) will be referred to as the data, while a
deformation /e Def that satisfies (3) and (4) will be called a solution corresponding to
the data (d, s, b).
Remark. The traction problem (Sf = d;M,Q) = 0) is excluded from our consideration
since stab(H) is, in general, zero when 2 is empty. For a partial resolution of this
problem cf. [5].
Remark. If s and b are continuous and / is a C2 solution corresponding to the data
(d, s, b) then the divergence theorem can be used to show that (3) and (4) are equivalent
to
div S(V/) 4- b = 0 in @t,
f = d on
S(Wf)n= s on y.
Note that the definition of stability is independent
depend on 3>, the domain of d.
Our main result is the following
of the data although it does
Theorem. Let Q be convex5 and H-stable. Then there exists a constant X > 0, which
depends only on stab(fi) and gft, such that if / and / are solutions that lie in Q and
correspond to data (d, s, b) and (d, s, 5) respectively, then
\\f —f\\mm) ^ A(||s—s||L2(y>)
+ II6—&||l2(»))Remark. A direct consequence of our theorem is uniqueness: there is at most one
solution corresponding to (d, s, b) in Q. On the other hand, there may be additional
solutions which do not lie in Q and may therefore not depend continuously on the data.
(For uniqueness under weaker hypotheses, cf. [1].)
Proof of the theorem. Let Q c Def be convex and H-stable. Let u=f—f
Then /=/
on S>, so that u = 0 on Q) and u is a variation; hence (3) yields
f Vu•[S(V/)
- S(V/)]
= f u ■(s- s)+ f u ■(S- b).
(5)
Consider the function g: [0, 1] -►U defined by
g{c)= f Vu • S(V/+ (7Vu).
By (1)
g'(a) = I Vu ■/4(V/+ a Vu) Vu.
Hence, by the mean value theorem, gf(l) —g(0) = g'(£) at some
e (0, 1); so that
[ Vu • [S(V/)- S(V/)]= [ Vu • A(Wf+I Vu)Vu.
* a*
'' Convexity here is with respect to the linear structure in C'(,
(6)
138
SCOTT J. SPECTOR
Since Q is convex, (/ + <*u)e t) and therefore (2) gives us
stab(n)||V«||21(a)
< f Vu■A(Vf+HVw)V«.
Next, by the Cauchy-Schwarz
(7)
inequality,
(" u ■(b — b) < \\u\\L2m || fi — b||t2(s»),
^
,
I U ■(s — s) < ||w||t2(^) ||s —s||/_2(y)•
V
(8)
JSf
Finally, the desired result follows from (5), (6), (7), (8) and the standard inequalities5
I|u||t2(<»)5S
||U||L2(y)
||V«||t2(a))
^ ^l(||U||£J(»)
+ ||Vu||
where
> 0 depends only on 38.
Remark. It is clear from the proof of the theorem that the L2-norm of (/ —/) can be
replaced by the H'-norm:
IMIhi<s9)
= |ImIIl2(s»)
+ ||^uIIl2(«)In addition, if we replace (8) by7
| u ■(b — b) < ||w||//i(<»)||—^||h-i(»)!
I U • (s — s) < ||u||//i/2(y) ||s — ■s||//-i/2(y)»
and use the trace theorem (cf. e.g. [6, p. 277])
IMIhi/2(y)^ ^2 ||M||//i(<ss)>
we arrive at the stronger result
||/ —/||«i(ag) ^ ^(||s ~ s||//-i/2(y) + IIS —b\\H-1(m)).
References
[1] M. E. Gurtin and S. J. Spector, On stability and uniqueness in finite elasticity, Arch. Rat. Mech. Anal., to
appear
[2] G. Fichera, Existence theorems in elasticity, in Handbuch der Physik, VIa/2, Berlin, Springer-Verlag (1972)
[3] J. Hadamard, Lemons sur la propagation des ondes et les equations d'hydrodynamique, Paris, Hermann (1903)
[4] S. J. Spector, On uniqueness in finite elasticity with general loading, J. Elasticity, to appear
[5] S. J. Spector, On uniqueness for the traction problem in finite elasticity, in preparation
[6] R. A. Adams, Sobolev spaces, New York, Academic Press (1975)
6 The first is the Poincare inequality, cf., e.g., Fichera [2, p. 274, footnote 17], whose proof with minor
modification applies in the present circumstance. The second is the trace theorem, cf., e.g., Fichera [2, p. 353],
Adams [6, p. 113].
7 The negative and fractional norms are defined in the usual manner. Cf., e.g., [6].